Identification and Classification of Acute Lymphocytic Leukemia Using Weighted Ensemble Neural Networks

Abstract

Hematologic malignancy puts the lives of numerous people of all the ages at serious risk globally. Acute lymphoblastic leukemia (ALL), a sub type of leukemia, is a dreadful disease that affects millions of individual globally. Early detection is crucial, and image analysis of peripheral blood smears is a simple way to diagnose and categorize leukemic cells. However, image analysis is complex and can lead to incorrect findings. To address this, a model is proposed to examine ALL and identify its various categories using deep ensemble learning. This study shows that a deep learning method based on an ensemble weight threshold improves leukemia detection using imbalanced dataset which reflects the real-life scenario. The proposed model examines healthy and ALL cells and identify its various categories, including Benign, Pro-B, Early Pre-B, and Pre-B from peripheral blood smear images to help in the early detection of this life-threatening disease with the highest accuracy (99.96%), precision (99.86%), sensitivity (99.93%), specificity (99.97%), F1-Score (99.89%). The model also outperforms five other pre-trained networks, viz. AlexNet, VGG-16, DenseNet-210, ResNet-50, and MobileNetV2 in classification accuracy, demonstrating its potential as an effective early diagnostic tool for ALL and its subtypes